Live API Keys Discovered in LLM Training Data Archive

# Live API Keys Discovered in LLM Training Data Archive

https://gcdradar.vercel.app/incidents/live-api-keys-discovered-in-llm-training-data-archive · AIR-2025-0171 · closed 2025-03-14

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Fix status unknown
- First reported: 2025-02-28
- AI role: Attack on AI
- Products: Common Crawl, Azure OpenAI Service

## Summary

Security researchers found thousands of live API keys and authentication credentials inside a web archive used for training large language models. The exposed secrets created risks for unauthorized access and downstream service compromises.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2025-02-28: Alleged Inclusion of 12,000 Live API Keys in LLM Training Data Reportedly Poses Security Risks (https://incidentdatabase.ai/cite/956)

## Sources

- 2025-02-28 AI Incident Database: Alleged Inclusion of 12,000 Live API Keys in LLM Training Data Reportedly Poses Security Risks https://incidentdatabase.ai/cite/956

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.